Why logging your workouts works, and what the evidence actually shows

Progressive overload is the principle that the demands on the body have to keep rising for adaptation to continue. The ACSM's 2009 position stand puts it plainly: systematically increasing the demands placed upon the body is necessary for further improvement. It then gives the working rule. When training at a given repetition-maximum load, add 2 to 10% when you can perform the current workload for one or two repetitions over the target on two consecutive training sessions. The lower end is for small-muscle exercises, the higher end for large ones. The ACSM grades this recommendation B, not A, so treat the percentages as convention and the structure as the point.

The structure is what matters here. "Two consecutive sessions" is a comparison across time. It requires you to know what load you used, how many reps you got, and whether that beat the target, on two separate days. Nobody remembers that reliably across six exercises and a fortnight. A log is the only honest way to apply the rule.

It also does not matter which variable you progress. Plotkin and colleagues randomised trained adults to eight weeks of lower-body training where one group added load and held reps constant, and the other added reps and held load constant. Both groups got stronger and both grew. Rectus femoris growth modestly favoured the reps group (2.8 mm, 90% CI −0.5 to 5.8), dynamic strength slightly favoured the load group (2.0 kg, 90% CI −2.4 to 7.8), and the authors call those differences of questionable practical significance. Either route works. Either route needs last session's number.

There is no study measuring how well lifters remember their previous loads without a record. The nearest evidence is about recalling physical activity in general, and it is not flattering. Prince and colleagues reviewed 187 studies comparing self-reported activity with direct measurement. Correlations were low to moderate, with a mean of 0.37. Sixty percent of the comparisons found people reported more activity than the instruments recorded, by a mean of 44% against accelerometers, and the error was largest for vigorous activity. That is questionnaire recall of walking and running, not recall of a bench press weight. Read it as a warning about memory, not as a measurement of yours.

The short answer — Write down the exercise, the load, the reps and how hard it was, at the time. Progress one variable when you beat the target twice in a row. That rule is the whole reason to keep the log.

The behaviour-change literature has tested self-monitoring more thoroughly than any other single technique. Harkin and colleagues pooled 138 randomised experiments and 19,951 participants in which people were prompted to monitor progress toward a goal. Monitoring raised goal attainment with an effect size of d = 0.40 (95% CI 0.32 to 0.48). Two moderators are directly relevant. Physically recording the information gave larger effects than monitoring without recording (d = 0.43 versus 0.29), and the gap was wider when attainment was measured objectively (0.57 versus 0.23). Reporting what you monitored to someone else beat keeping it private (0.47 versus 0.19). In the physical-activity subgroup specifically, the effect was medium at d = 0.59. A written diary, the closest thing in the review to a training log, came in at d = 0.42.

Two cautions. Almost every study in that review was about a health goal, and none was about lifting. The moderator comparisons are between studies, not randomised, so "recording causes the extra effect" is a reasonable reading rather than a proven one.

Michie and colleagues asked a different question of 122 diet and physical-activity interventions covering 44,747 people: which technique explains the most variation in results? Self-monitoring did, at 13% of the heterogeneity. Interventions that combined self-monitoring with at least one other self-regulation technique, such as goal setting, feedback or reviewing goals, had an effect of 0.42 against 0.26 for the rest. That is the shape of a training log used well: a record, plus a target, plus a look back.

The wearable-tracker literature makes the same point at scale, with a caveat about what was measured. An umbrella review by Ferguson and colleagues covered 39 systematic reviews and about 164,000 participants and found trackers raised physical activity by an SMD of 0.3 to 0.6, roughly 1,800 extra steps and 40 more minutes of walking a day, with gains durable for at least six months. Laranjo's meta-analysis of 28 randomised trials in healthy adults found a similar standardised effect and found that personalisation and messaging made interventions more effective. Chaudhry's review of 57 step-count trials shows the effect fades: about 1,126 extra steps a day within four months, 464 at one year. All of that is steps, not sets. It shows monitoring changes behaviour. It does not show what it does to a deadlift.

Load and reps are not the whole session. The same set of five at 100 kg means different things at RPE 7 and RPE 10, and only one of them means you are out of room. Logging effort matters because effort is the number lifters get wrong most predictably.

Halperin and colleagues meta-analysed 12 studies and 414 participants who were asked how many reps they had left. On average people under-predicted by 0.95 reps (95% CI 0.17 to 1.73), meaning they stopped with more in the tank than they thought. Training status made no difference to accuracy in that analysis. Hackett's study of 81 lifters gives the shape: the error is about one rep when you are within five reps of failure and more than two reps when you are seven to ten reps out. At the other extreme, Zourdos found that novices rated a true one-rep max at RPE 8.96 on average, against 9.80 for experienced squatters (p = 0.023). Beginners literally do not know what a maximum feels like.

An effort score written down next to the load is how you find that out about yourself. It is also how the log stays honest when the numbers stall: the same reps at the same weight at a higher RPE is a different result from the same reps at the same RPE, and only a record shows it. Our guide to RPE and RIR covers how to rate a set.

There is a further reason to log how the session felt, from athlete monitoring. Saw, Main and Gastin reviewed 56 studies comparing self-reported wellbeing with objective markers such as heart rate, hormones and blood measures. Subjective measures reflected changes in training load with better sensitivity and consistency than the objective ones, and a single blended score was less sensitive than the individual items. A line about sleep or soreness next to your sets is not fluff. In athletes it outperformed the lab.

Some things only exist across sessions. A stall is one of them. As our guide on stalls argues, one flat session is noise, and the signal is fewer reps at the same load at the same or higher effort across two or three weeks. You cannot see a trend at a fixed load without the fixed load being written down.

Training-load monitoring in sport runs on the same idea. Foster's session-RPE method, from 1998, has athletes record one effort rating for the whole session and multiply it by duration. His original study of 25 athletes found illnesses clustered when individuals crossed their own thresholds of training strain, and the thresholds were individual, which is itself an argument for a personal record.

One popular extension deserves a warning. The acute-to-chronic workload ratio, proposed by Gabbett in 2016, compares this week's load with the previous four and claims a "sweet spot" between 0.8 and 1.3 and a "danger zone" above 1.5. It has since been strongly criticised. Impellizzeri and colleagues showed that an additive model was transformed into a ratio without justification, and in a later paper found that dividing by a randomly generated denominator reproduced the same injury association. Log your training. Do not buy an app for its injury-risk ratio.

The gap in this literature is specific and worth stating. No randomised trial has assigned lifters to log or not log their sets and measured strength, size or adherence. A 2025 review of 32 randomised trials of mobile-health interventions for resistance training found only two that measured changes in resistance-training participation at all, too few to pool.

The nearest trials cut both ways. In the Phys-Can trial, 577 people undergoing cancer treatment did supervised resistance training with or without coached goal setting, self-monitoring and planning. Attendance was 52% of prescribed sessions with the techniques and 53% without, a difference of −0.6 percentage points (95% CI −5.6 to 4.4). But every participant in both arms kept a training log, so the trial tested coaching on top of logging, not logging itself. In Gavanda's trial of 79 trained adults on an identical ten-week programme, 81.2% of those given a tracking app met the 85% attendance threshold against 52.2% given a static PDF. The app bundled video demonstrations, progress tracking and automated reminders, the paper reports no significance test on that gap, and the analysis excluded people below the threshold. It is suggestive, not proof.

One more finding argues for humility. Weakley's group gave twelve rugby players live bar-velocity readouts during a set of squats, either shown on a screen or read aloud, or plain verbal encouragement with no numbers at all. All three raised velocity by a similar amount over no feedback. The information was not obviously the active ingredient. Real-time feedback is not the same thing as reading last week's log, but the result is a reminder that "seeing the number" is not magic.

• Exercise, load, reps, for every working set. This is the input the ACSM rule runs on.
• Effort, as RPE or reps in reserve. It is the number you are worst at estimating and the one that makes a flat week readable.
• One line on how it felt. Sleep, soreness, a niggle. Per-item self-report beat objective markers in athletes.
• Compare across sessions, not within one. Progress one variable when you have beaten the target twice. Change one thing at a time.

The direct evidence that logging improves resistance-training outcomes does not exist yet. The progression argument rests on a position-stand rule graded B and on a single eight-week trial showing load and rep progression both work. The self-monitoring effects come from health goals and step counts, in populations that mostly were not lifters, and the moderators inside those reviews are between-study comparisons. The effort-accuracy findings are robust but say nothing about whether writing RPE down improves it. The one trial that came close to testing an app in trained lifters bundled three features and ran no significance test. What is left is a strong mechanistic case and a consistent behavioural one, and the missing trial should be named as missing.

Shojin is built around the two-session rule. When you open an exercise, last session's load and reps are already filled in, so the comparison the ACSM rule needs is on screen before your first set. Every set takes an RPE, and the progression suggestion reads it: a top set at RPE 8 or under suggests one increment up, RPE 9 to 10 holds, and reps falling twice at RPE 9 or above backs the load off by about 5%. Personal records and estimated one-rep max are computed from the whole history, so a record is never something you have to notice yourself. Weekly working sets per muscle come from the same log, measured against evidence-based targets on our methodology page.

Shojin Coach writes your next block from that history. It does not compute an acute-to-chronic workload ratio, and it does not track sleep or food, so the recovery question above is one it cannot answer for you.

Every number in this article traces to one of these. Where the evidence is contested or thin, the article says so rather than picking a side.